Humanoid Robots Self-Train on Real Hardware
Scobleizer · x · 2026-07-11
Humanoid robots are moving beyond merely imitating human demonstrations; they are now continuously learning via reinforcement learning directly on real-world production tasks.
The post notes that Kinetiq Ascend allows robots to train 24/7 on real hardware, learning from their own successes and failures rather than just copying demos:
- In pick-and-place and handoff tasks, the success rate jumped from 80% to 98%, with failures dropping about 10x
- In bimanual bin handling tasks, throughput more than doubled with a success rate nearing 99%
- These improvements were achieved in "days of robot time," not months
The core takeaway is that reliable, general-purpose humanoid robots will scale and improve continuously with compute and data, much like LLMs. Every deployed robot will contribute to future training data.
Related event: Humanoid Robots Shift to Real-World Reinforcement Learning(3 posts)→
More from Embodied
- Humanoid robots are moving from labs into public culture — Olivier__OG · 2026-07-21
- Polymarket puts Tesla’s California robotaxi launch odds at 16% this year — Polymarket · 2026-07-21
- Tesla expands robotaxi service to Orlando and Tampa — Polymarket · 2026-07-21
- Humanoid robots are approaching a deeper uncanny valley — GlenBradley · 2026-07-21
- Polymarket gives Tesla’s Optimus just a 17% chance of debuting this year — Polymarket · 2026-07-21
- UK robotics startup Humanoid raises $152 million at a $1.35 billion valuation — Polymarket · 2026-07-21